potential performance issues

[BurntSushi]

I haven't done any benchmarks, but from reading the code, I see two potentially quite large performance problems.

1. You're cloning the entire RegexSet on each call to handle. It's true that cloning a Regex or a RegexSet is in and of itself very cheap, but in doing so you will have lost any state built up from previous matching, which will impact performance. (One might argue that this is a problem with Regex and that it should clone the internal state, but then your clone won't be cheap.)
2. In handle, the code is compiling a Regex for every match. Regex compilation is not something you should think of as being fast, and in this case, will probably negate many of the perf benefits. Instead, I'd recommend compiling each regex added to RegexSet as it is built, and then using the index you get back from matches to find the right regex to use.

[gsquire]

You're absolutely right, I will have to revisit that part of the code. I can add that state to the structure so lookups can be faster. Thanks for taking the time to point out some flaws!

[gsquire]

So I managed to triple requests per second after this commit with my simple wrk benchmark. It was around 85K which I was impressed with. I always take those with a grain of salt, but that's much better than before! See here: 6db7c26

[gsquire]

Also, thank you for doing another CR. I really appreciate it.

[BurntSushi]

@gsquire Looks much better! Nice work. :-)